HELM VHELM - Blink Perception - Relative Depth: leaderboard
Metric: Prefix Quasi-Exact Match (%). Source: crfm.stanford.edu. 48 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Llama 4 Scout Instruct | 96.77 |
| 2 | Llama 4 Maverick Instruct FP8 | 91.94 |
| 3 | Gemini 2.5 Pro (Preview 03-25) | 90.32 |
| 4 | Claude 3 Opus (20240229) | 89.52 |
| 5 | GPT-4.5 | 88.71 |
| 6 | GPT-4.1 (2025-04-14) | 87.1 |
| 7 | Gemini 2.0 Flash Lite | 87.1 |
| 8 | O3 (2025-04-16) | 86.29 |
| 9 | Gemini 1.5 Pro (002) | 84.68 |
| 10 | Gemini 1.5 Flash (002) | 83.87 |
| 11 | Gemini 2.0 Pro (Preview 02-05) | 83.87 |
| 12 | Gemini 2.0 Flash | 83.06 |
| 13 | O4 Mini (2025-04-16) | 83.06 |
| 14 | Claude 3 Sonnet (20240229) | 82.26 |
| 15 | GPT-4.1 Mini | 81.45 |
Interactive version: theaggregate.ai/benchmark?slug=helm-vhelm-blink-perception-relative-depth · How It Works · Data refreshed daily, snapshot 2026-09-19.